January 2021 Hardware Upgrades

[Jim]

Turns out there was a misunderstanding on the memory, and it has not yet arrived.

When the new memory arrives and noreaster has to go offline, I don't suppose you could humor me and do a round of memtest86 with the existing DIMMs?

Can probably make that happen.

[Jim]

I'm finally planning to complete the hardware upgrades by doubling the server's RAM on the afternoon of Tuesday, March 16.  Down time shouldn't last long, but TM, the forum and METAL will all be down during the upgrade.

[Jim]

Heading to the machine room shortly for the memory upgrade.  Expect several minutes to maybe a half hour of down time between 1:00 and 2:00 PM EDT today.

[Jim]

What was supposed to be a few minutes of downtime to pop in the new memory chips turned into about an hour tracking down the fact that 2 of the 8 memory chips were a slightly different part and not compatible with the other memory in the system.  The good news is that the system at least went from 32 GB of RAM to 56 GB, though not the 64 GB it was intended to be.  Everything should now be back to normal.  There will be a hopefully shorter outage whenever the correct chips finally arrive.

[yakra]

Oh fun. Sounds like you're not getting the full bandwidth benefit of n channels then?

[Jim]

Oh fun. Sounds like you're not getting the full bandwidth benefit of n channels then?

Probably not.  Hopefully the two misfits will be replaced fairly quickly.

[yakra]

When the new memory arrives and noreaster has to go offline, I don't suppose you could humor me and do a round of memtest86 with the existing DIMMs?

Can probably make that happen.

Did this happen?
The mysterious C++ Pointer Value Loses 0x900000000 bug is back.
It's weird enough that I can't write off bad memory as a potential cause.

[Jim]

When the new memory arrives and noreaster has to go offline, I don't suppose you could humor me and do a round of memtest86 with the existing DIMMs?

Can probably make that happen.

Did this happen?
The mysterious C++ Pointer Value Loses 0x900000000 bug is back.
It's weird enough that I can't write off bad memory as a potential cause.

We didn't do any additional testing.  I still haven't acquired the 2 additional ones to get up to the full 64 GB.

[yakra]

The Bad RAM Hypothesis is ruled out.

[Jim]

I have probably mentioned this before, but I spent a ridiculous amount of time in grad school tracing a bug that wound up being a bug in IBM's compiler, not my code...

[yakra]

Haha, I've run into one compiler bug already. At least that time I had something to Google; this time not so much.
Tracking down whatever this is ain't gonna be pretty.
I guess one relatively low-effort thing I can try is to install FreeBSD 13.0 and see if there's a new version of clang & things behave differently...

[yakra]

Still waiting on those two DIMMs?

[Jim]

Still waiting on those two DIMMs?

They were unavailable for a while and I never pursued it further.  Maybe I still should.